Outbound calls fail from the whole cluster whenever traffic peaks

• During peak hours, services across all nodes see sporadic connect: connection timed out to external APIs.
• The same APIs answer instantly from a laptop outside the cluster.
• Failures correlate with request volume, not with any particular destination.
• Some connections are reset mid-request rather than failing to connect.

# on a worker node (Linux)
$ dmesg | tail -3
[812031.117] nf_conntrack: nf_conntrack: table full, dropping packet
[812031.121] nf_conntrack: nf_conntrack: table full, dropping packet

$ sysctl net.netfilter.nf_conntrack_max net.netfilter.nf_conntrack_count
net.netfilter.nf_conntrack_max = 65536
net.netfilter.nf_conntrack_count = 65536

$ conntrack -L 2>/dev/null | awk '{print $4}' | sort | uniq -c | sort -rn | head -3
  52210 TIME_WAIT
   9105 ESTABLISHED
   3811 SYN_SENT

$ sysctl net.netfilter.nf_conntrack_tcp_timeout_time_wait
net.netfilter.nf_conntrack_tcp_timeout_time_wait = 120

# the cloud NAT gateway metrics for the same window
ErrorPortAllocation   spikes to 1,400/min at 17:00–17:20
ConnectionAttemptCount ~ 9,800/s across the cluster